Sampling and estimation techniques for surveillance and monitoring

Abstract

The complexity of issues facing animal agriculture and veterinary and human medicine necessitate the use of appropriate scientific methods for obtaining and analyzing data so that objective inferences can be made and practical solutions can be sought. An overall objective of these studies was to examine the validity of underlying assumptions of sampling methodologies in the context of antimicrobial susceptibility and surveillance and monitoring for animal diseases. A secondary objective was to examine analytical methods that can be used to address the intricacies of antimicrobial susceptibility data. Evaluation of sampling strategies for measuring antimicrobial susceptibility patterns in isolates taken from the feces of feedlot cattle was the subject of two studies. In the first study, rectal and pen floor fecal samples were collected. Individual samples were used to create pools of 5 and 10 fecal samples. Five Escherichia coli isolates were obtained from each individual and pooled sample. The susceptibility patterns were compared among the collection methods and the individual/pooled samples. Antimicrobial resistance patterns from isolates obtained from rectal samples did not differ from isolates obtained from pen floor fecal samples. Little pen-to-pen variation in resistance prevalence was observed but clustering of resistance phenotypes within pens and samples was detected. Pooling of fecal samples yielded resistance patterns that were consistent with those of single fecal samples when the prevalence of resistance to an antimicrobial was > 2 percent. Pooling may be a practical alternative when investigating patterns of resistance that are not rare. The second study of antimicrobial resistance in E. coli isolated from feedlot cattle focused on the short-term repeatability of antimicrobial susceptibility patterns. Fecal samples were collected from the floors of six pens on two sampling occasions separated by 48 hours. Resistance to individual antimicrobials was consistent across the periods and individual/pooled samples by all analytical measures when the prevalence of resistance was at least 2 percent. Inconsistent results were obtained for antimicrobials to which resistance rarely occurred. The apparent inconsistencies did not appear to be related to external factors but rather to sampling intensity. Short-term stability is a plausible assumption under sampling strategies designed to detect specific prevalence levels. However, when resistance levels are low there likely will be fluctuations in the occurrence, prevalence and central tendency measures of rare resistance phenotypes. Factor analysis was used to explore resistance and susceptibility patterns of the minimum inhibitory concentration for the 17 antimicrobials tested on is. coli isolates. New generation cephalosporins, older generation beta-lactams, fluoroquinolones and aminoglycosides grouped separately as classes of antimicrobials on four of the six factors. One of the remaining factors was a grouping of antimicrobials that had been identified as being related in previous feedlot studies. The last factor was a grouping of three of the five antimicrobials that comprise the antimicrobials found in penta-resistant strains of Salmonella Typhimurium. The factor analysis provided patterns in the MIC data that would not have been apparent if the antimicrobial-resistance data had been analyzed merely according to the susceptible/resistance categories. Two-stage sampling designs are appropriate for disease surveys. A sample size formula for estimating herd-level prevalence was proposed that depends on herd-level sensitivity and specificity. The impact of the distribution of the within-herd prevalence as determined by animal-level prevalence and the intracluster correlation coefficient was modeled using a Monte Carlo simulation. At low prevalence, herd-level sensitivity increased with increasing intracluster correlation, but sensitivity was less affected at higher prevalence. Also at low prevalence, many herds were being classified as positive based only on false positive test results. Positive predictive values dropped sharply with increasing intracluster correlation. A hypothetical and two real life two-stage sampling designs were used as examples to evaluate the model and the sample size formula. The use of a distribution for within-herd prevalence resulted in a conservative estimate of herd-level test characteristics. The model allows researchers to trade off between the number of herds and the number of animals sampled by manipulating herd-level test characteristics.